I have a feeling we are paying for massive VMs that only use 5% of their CPU. How do I get an actual recommendation from Azure on resizing? I have looked at Azure Advisor, but is it accurate? I want to know if there is a way to see real-time usage data to make an informed decision about downsizing.
To determine proper VM sizing, users should leverage Azure Monitor and Log Analytics to analyze p95 performance metrics over a sustained period rather than relying solely on Advisor recommendations.
4 answers
Azure Advisor provides a decent surface-level scan, but you need to configure custom metrics for real visibility.
- Install the Log Analytics agent on your target VMs to gather granular performance data
- Query the InsightsMetrics table in your Log Analytics workspace to identify true CPU peaks
- Compare those peaks against your current SKU capacity to validate if the downsizing is safe
Azure Advisor is a baseline tool, but it often misses the cyclical workload peaks that lead to project slippage. You should rely on Azure Monitor metrics to verify sustained low utilization before committing to a SKU downgrade.
I recall auditing a client portfolio where we identified a 30 percent waste by ignoring the difference between average and p95 CPU utilization. We pulled the metrics directly from the Azure Monitor logs instead of trusting the high-level Advisor dashboard.
By mapping that historical data against our actual financial burn rate, it became obvious that the VMs were drastically over-provisioned. We successfully downsized without impacting service performance once we had the granular trend lines to justify the shift to leadership.
Advisor is helpful for a quick check, but it remains a blunt instrument compared to the precision of deep-dive performance analysis. Relying solely on automated recommendations works if your workloads are entirely predictable, but it is better to perform a trend analysis when you are managing cost-sensitive infrastructure.
If your team requires absolute stability, the extra cost of over-provisioning acts as a safety buffer, yet that buffer is usually inefficient for standard production environments. When you prioritize fiscal discipline, you must align your actual consumption data with the cost-to-performance ratio of smaller instance families.
